Model tiering for IoT device clusters

Abstract

Edge devices of a network collect data. An edge device may determine whether to process the data using a local data processing model or to send the data to a tier device. The tier device may receive the data from the edge device and determine whether to process the data using a higher tier data processing model of the tier device. If the tier device determines to process the data, then the tier device processes the data using the higher tier data processing model, generates a result based on the processing, and sends the result to an endpoint (e.g., back to the edge device, to another tier device, or to a control device). If the tier device determines not to process the data, then the tier device may send the data on to another tier device for processing by another higher tier model.

What is claimed is: 
     
       1. A system, comprising:
 one or more edge devices connected to a network, wherein individual edge devices of the one or more edge devices are configured to: 
 collect data by the edge device as current data;
 determine by the edge device that collected the current data, based on whether an amount of energy that would be required for a data processing model of the edge device to process the current data collected by the edge device is above a threshold amount of energy and also based on a confidence level of one or more previous predictions that were previously generated by a data processing model of the same edge device by processing different data that was previously collected by the same edge device before the current data was collected and wherein the data processing model is trained based on the same training data that the data processing model was previously trained on to generate the previous predictions, whether to:
 process the current data at the same edge device by using the same data processing model trained based on the same training data that the data processing model was previously trained on to generate the previous predictions in order to generate a new prediction at the edge device based on the current data collected by the edge device, or 
 send the current data collected by the edge device to one or more tier devices connected to the same network; and 
 
 in response to the determination of whether to process the current data at the same edge device or to send the current data to the one or more tier devices:
 process the current data using the data processing model of the edge device, or 
 send the current data to the one or more tier devices; and 
 
 
 the one or more tier devices, wherein individual ones of the tier devices are configured to:
 receive the current data from the edge device; 
 determine whether to process the current data using a higher tier data processing model of the tier device; and 
 in response to determining to process the current data using the higher tier data processing model:
 process the current data using the higher tier data processing model; 
 generate a result based on the processing; and 
 send the result to an endpoint. 
 
 
 
     
     
       2. The system as recited in  claim 1 , wherein the higher tier data processing model of the tier device is larger than the data processing model of the edge device, and wherein the result is another prediction, and wherein the higher tier data processing model is configured to:
 generate the other prediction with a higher confidence level than the prediction generated by the data processing model of the edge device. 
 
     
     
       3. The system as recited in  claim 1 , wherein the edge device is configured to:
 determine, based on one or more additional criteria, whether to process the current data using the data processing model of the edge device or to send the current data to the tier device, wherein the one or more additional criteria comprises one or more of:
 an amount of time required for the data processing model to process the current data, 
 an amount of time since the data processing model was updated, and 
 an amount of bandwidth available to send the current data to the tier device. 
 
 
     
     
       4. The system as recited in  claim 1 , wherein the edge device is configured to:
 generate the prediction based on processing of the current data using the data processing model of the edge device; 
 determine that a confidence level of the prediction is below a threshold confidence level; and 
 in response to the determining that the confidence level of the prediction is below the threshold confidence level, send the current data that was used by the edge device to generate the prediction to the one or more tier devices. 
 
     
     
       5. The system as recited in  claim 1 , wherein the tier device is configured to:
 in response to determining not to process the current data using the higher tier data processing model:
 send the current data to another tier device or to a provider network; 
 receive a result from the other tier device or the provider network, wherein the result is based on the current data; and 
 send the result to the edge device. 
 
 
     
     
       6. A method, comprising:
 one or more tier devices connected to a network that comprise a respective data processing model, wherein individual tier devices of the one or more tier devices perform:
 receiving data by the tier device as current data; 
 determining by the tier device that received the current data, based on whether an amount of energy that would be required for a data processing model of the edge device to process the current data collected by the edge device is above a threshold amount of energy and also based on a confidence level of one or more previous predictions that were previously generated by a data processing model of the same tier device by processing different data that was previously collected by the same tier device before the current data was received by the tier device and wherein the data processing model is trained based on the same training data that the data processing model was previously trained on to generate the previous predictions, whether to: 
 process the current data at the same tier device by using the same data processing model trained based on the same training data that the data processing model was previously trained on to generate the previous predictions in order to generate a new prediction at the tier device based on the current data collected by the tier device, or 
 send the current data collected by the tier device to one or more other tier devices connected to the same network; and 
 
 in response to the determining:
 processing the current data using the data processing model of the tier device to generate the prediction, or 
 sending the current data to the one or more other tier devices. 
 
 
     
     
       7. The method as recited in  claim 6 , further comprising determining whether to process the received data at the tier device or to send the current data to one or more other tier devices connected to the network based on an amount of time since the data processing model was updated. 
     
     
       8. The method as recited in  claim 6 , wherein another data processing model of the one or more other tier devices is larger than the data processing model of the tier device. 
     
     
       9. The method as recited in  claim 8 , further comprising determining whether to process the current data at the tier device based on a confidence level of one or more respective predictions previously generated by another data processing model of the one or more other tier devices. 
     
     
       10. The method as recited in  claim 6 , wherein the current data comprises a plurality of portions of data, and wherein the sending the current data comprises:
 sending respective portions of the current data to respective tier devices of a plurality of the other tier devices; and 
 receiving one or more results from one or more of the other tier devices, wherein the one or more results are based on the respective portions of the current data. 
 
     
     
       11. The method as recited in  claim 6 , wherein the received data comprises data that has been transformed by another data processing model of another tier device. 
     
     
       12. The method as recited in  claim 6 , wherein sending the current data to the one or more other tier devices comprises sending the current data to a remote provider network. 
     
     
       13. The method as recited in  claim 6 , wherein the receiving data comprises:
 receiving the current data from one or more environmental sensors of the tier device. 
 
     
     
       14. A non-transitory computer-readable storage medium storing program instructions that, when executed by a tier device connected to a network, cause the tier device to implement:
 receiving data by the tier device as current data; 
 determining by the tier device that received the current data, based on whether an amount of energy that would be required for a data processing model of the edge device to process the current data collected by the edge device is above a threshold amount of energy and also based on a confidence level of one or more previous predictions that were previously generated by a data processing model of the same tier device by processing different data that was previously collected by the same tier device before the current data was received by the tier device and wherein the data processing model is trained based on the same training data that the data processing model was previously trained on to generate the previous predictions, whether to:
 process the current data at the same tier device by using the same data processing model trained based on the same training data that the data processing model was previously trained on to generate the previous predictions in order to generate a new prediction at the tier device based on the current data collected by the tier device, or 
 send the current data collected by the tier device to one or more other tier devices connected to the same network; and 
 
 in response to the determining:
 processing the current data using the data processing model of the tier device to generate the prediction, or 
 sending the current data to the one or more other tier devices. 
 
 
     
     
       15. The non-transitory, computer-readable storage medium of  claim 14 , further comprising program instructions that cause the tier device to implement determining whether to process the received data at the tier device based on one or more of:
 an amount of time required for the data processing model to process the current data, or 
 an amount of time since the data processing model was updated. 
 
     
     
       16. The non-transitory, computer-readable storage medium of  claim 14 , wherein another data processing model of the one or more other tier devices is larger than the data processing model of the tier device. 
     
     
       17. The non-transitory, computer-readable storage medium of  claim 16 , further comprising program instructions that cause the tier device to implement determining whether to process the received data at the tier device based on one or more of:
 an amount of bandwidth available to send the current data from the tier device to the one or more other tier devices, 
 an amount of time required for the other tier data processing model to process the current data, 
 an amount of time since the other tier data processing model was updated, and 
 a confidence level of one or more respective predictions previously generated by the other tier data processing model. 
 
     
     
       18. The non-transitory, computer-readable storage medium of  claim 14 , wherein the current data comprises a plurality of portions of data, and wherein the program instructions cause the tier device to further implement:
 sending a respective portion of the current data to a respective tier device of a plurality of the other tier devices; and 
 receiving one or more results from one or more of the other tier devices, wherein the one or more results are based on the respective portions of the current data. 
 
     
     
       19. The non-transitory, computer-readable storage medium of  claim 14 , wherein the program instructions cause the tier device to further implement:
 receiving the current data from one or more environmental sensors of the tier device. 
 
     
     
       20. The non-transitory, computer-readable storage medium of  claim 14 , wherein the program instructions cause the tier device to further implement:
 sending the current data to a remote provider network.